SELF-FUNDED 3.5-YEAR PHD – Automatic segmentation of femur using deep learning combined with phantomless calibration for rapid personalised fracture risk predictions in clinical applications

Details

We are excited to announce a new PhD opportunity in the areas of modelling, biomechanics & deep learning. The project is in collaboration with multiple NHS Trusts including the Sheffield Teaching Hospitals and Sheffield Children’s Hospital. The successful candidate will join a collaborative and world-leading team of researchers within the Intergrated MusculoSkeletal Biomechanics group and the Insigneo Institute for in silico Medicine in Sheffield.

 Predicting the risk of fracture in long bone such as the femur provides important quantitative diagnostic and prognostic information in clinical practice, both in the application of predicting osteoporotic fracture in the elderly, the effect of prostate cancer treatments, and the diagnosis of non-accidental fractures in young children. The current pythonised Computed Tomography to Strength (pyCT2S) pipeline can rapidly generate a personalised finite element mesh based on bone segmentation from CT scans and perform simulations under personalised loading conditions to generate fracture risk predictions. To achieve real-time operation, the pyCT2S needs to be further developed to incorporate automatic segmentation of long bones (femora) from CT scans and a phantomless calibration algorithm for bone material property estimation that accounts for adaptive current in clinical CT scan protocols. This project will therefore use deep learning algorithms to automatically segment femora from CT scans and adapt the pyCT2S pipeline to achieve full automation. The project will bring a step change to enhance our capability to achieve real-time fracture risk predictions in the clinic.

 This project will use anonymised CT scans obtained at multiple NHS centres including the Sheffield Teaching Hospitals (elderly adults) and Sheffield Children’s Hospital (children). Most of these CT scans have already been segmented (e.g., labelled), although further segmentations (of the contralateral part) will need to be carried out as part of the project to increase the size of the dataset. Two distinctive datasets of labelled femora will be generated from this work, one of elderly adults and the other of children 0-16 years old. The phantomless calibration method will first be developed using the exiting MultiSim dataset as a proof of concept, before testing it in a larger clinical dataset using scans performed with adaptive current. These lead to two main objectives of this project:

1.    To train deep learning algorithms to automatically segment the proximal femur from individual CT scans.

2.    To investigate the use of phantomless calibration for clinical CT scans with adapative currents.

 If you are interested in applying and wish to discuss any details of the project informally, please contact the main supervisor Dr Xinshan (Shannon) Li at [email protected].

 Requirements:

Education A very good 4-year degree or Master’s degree in Biomedical Engineering, Mechanical Engineering, Bioengineering, General Engineering, Applied Mathematics or Physics (at least a UK 2:1 honours degree, or its international equivalent).

Knowledge, skills Strong programming skills in Python, machine learning, advanced finite element analysis; Desirable: computer modelling, solid mechanics, experience with Ansys or equivalent software, a final year project on a machine learning/biomechanics problem.

 Other requirements:

We are looking for a motivated and driven individual with a strong passion for machine learning, biomechanics and in silico medicine research. Excellent communication and scientific writing skills are also required as well as a strong ability to work both individually and in a team.

Deadline: as soon as possible or until position is filled.